Professional Documents
Culture Documents
Copyright (c) 2012 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 2, No 1, March 2012
ISSN (Online): 1694-0814
www.IJCSI.org 230
Boltz et al. [4] have proposed to define the energy as Thus all the patches are transformed to transform
a function of (an estimation of) the MCE distribution. domain subsequently the Euclidean distance of every
This function was chosen to be a continuous version patches of consequent frames and their total mean are
of the Ahmad-Lin entropy approximation, the calculated as follows.
purposed being to be more to outliers inherently
presented in the MCE.
L ; Pi
3. Dynamic and static objects segmentation by
means of efficient object extraction
i 1; k 1
(Ti ,k Ti 1,k ) 2
Copyright (c) 2012 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 2, No 1, March 2012
ISSN (Online): 1694-0814
www.IJCSI.org 231
Partitioning of the video sequence by detecting the The prior research [] of this author used the Fuzzy K-
scene changes is essential for characterization and means clustering for object extraction. The main idea
categorization of video. Basically, there are two behind the fuzzy K-means is the minimization of
types of scene changes, abrupt change and gradual objective function, which is normally chosen to be
change. Abrupt change or camera changes are the total distance between all patterns from their
apparently easy to detect because the difference in respective cluster centers. The distribution of objects
image properties between two consecutive frames is among clusters and the updating of cluster centers are
expected to be large. But the detection of changes in the two main steps of the K-means algorithm. The
gradual transition is very difficult because the algorithm alternates between these two steps until the
changes occurred over period of frames. To segment value of the objective function cannot be reduced
the static and dynamic objects in the environment of anymore. Given the frame
abrupt and gradual changes, the object detection is a
crucial process. The proposed segmentation fˆaj a
technique uses fuzzy c means algorithm for detect the . (4)
objects in every frame. Initially the objects in every
frame are identified for segmentation. Let K-means determine the c cluster Vi=1.C centers by
minimizing the objective function
{ a | 1 a A } be the result of shot
segmentation where ‘A’ is total no of shots and C n
Min j ( ,V ) ik f aj ik Vi
2
a { f aj | 1 a A ;1 j | a | } be the set of
i 1 k 1 (5)
similar shots where ‘ | a | ’ are the total no of frames
in ath shot in the segmented results. The initial frames ik [0,1] i , k and . is the Euclidean
in every shot are taken as key frame for object Where
distance measure .
extraction for example the f 11 is key frame for shot
1 which is known as f key (1) . Like wise each shot
having its own key frames. Let 3.2.2. Fuzzy C-Means Clustering
Fkey { f key(i ) 1 i A}
be the key frames set of The fuzzy C-means clustering do the fuzzy partition
every ith shot. The objects in every frame are rather than the hard partition, by using the objective
identified using fuzzy c means algorithm. The FCM function given in eq (7). This algorithm is proposed
is developed by Dunn in 1973 and improved by as an improvement to the fuzzy k-means clustering.
Bezdek in 1981. The value of ‘C’ is calculated using The FCM partition the collection of ‘n’ vector into C
the 3D histogram value. The 3D color histogram is groups and finds a cluster center in each group such
calculated for every frame to find the no of objects in that a cost function of dissimilarity measure is
the frame. The number of peak values in the 3D color The FCM algorithm assigns pixels to each
histogram determines the total number of objects in a minimized.
frame which is given as the value of c in FCM. category using fuzzy membership functions. Let fˆaj
Finally the clustering process yields the no of objects
in a frame. The overlapping objects are also be the frame which is to be clustered into ‘C’
identified using fuzzy c means clustering. clusters. The algorithm is an iterative optimization
that minimizes the cost function defined as follows.
mn C
M ikm f ajik Vi
'
1 m (6)
k 1 i 1
Copyright (c) 2012 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 2, No 1, March 2012
ISSN (Online): 1694-0814
www.IJCSI.org 232
G G1 G 2 (10)
Copyright (c) 2012 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 2, No 1, March 2012
ISSN (Online): 1694-0814
www.IJCSI.org 233
The segmented motion objects are subjected Our proposed video segmentation approach has been
to dual morphological operations for getting the validated by experimenting with variety of video
improved segmentation results. Morphological sequences. The proposed system has been
processing is constructed with operations on sets of implemented in Matlab (Matlab7.10). The
pixels. The four most basic operations in performance of the proposed approach is compared
mathematical operations are dilation, erosion, with the results of the prior research of the author.
opening and closing.
(a)
(b)
Figure 1: (a) Sample frames, (b) Grayscale images,
The fig. 1(a) represents the sample frames(Frame :2,Frame :3, Frame :4, Frame :5, Frame :44) in a tested video and
the fig. 1(b) represents their corresponding gray scale converted image .Considering the initial frame as key frame
all the objects in the every frames are extracted. The proposed hybrid segmentation technique yields the dynamic
object segmentation results by intersection of segmented results of both the frame difference algorithm and
intersection methods and hence produces the better enhanced segmented results. Also the proposed system segments
the static objects in every frame.
The performance of the proposed system is evaluated (non-segmenting the objects). The output of the
by the statistical measures like sensitivity and proposed system may or may not be match with the
specificity. The output of the proposed system may original status of the image. Consider the following
be positive (Segmenting the objects) or negative setting for the statistical measures.
True Positive (TP) : Valid objects correctly segmented.
Copyright (c) 2012 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 2, No 1, March 2012
ISSN (Online): 1694-0814
www.IJCSI.org 234
4.2 Comparative Analysis comparing its segmented results with that of the
traditional video segmentation technique which uses
The performance of the proposed hybrid background substraction method.
segmentation technique is also evaluated by
Figure 2: Segmented static and dynamic objects(prior research)
Figure 3: Segmented static and dynamic objects (proposed method)
The fig. 2 represents the video segmentation results clustering) of the author. The table-1 and table-2
of the frames (Frame: 3, Frame: 4, Frame: 5 and represents the comparison statistical measures of the
Frame: 7) using the prior research (using FKM segmentation of video-I using the proposed technique
clustering) and fig.3 represents the segmentation as well as the prior method.
results using the proposed research (using FCM
Table1: Statistical Measures of the segmented results of Video‐I using the prior research
Mathews Kappa
Positive Negative False Correlation Coefficient
Measures TP TN FP FN Accuracy Predictive Predictive discovery Coefficient
value value(NPV) rate
Frame1 5270 92966 3140 1756 95.25 62.66 98.15 37.34 0.66 0.94149
Frame2 5814 93355 2207 2841 95.16 72.48 97.05 27.52 0.67 0.957513
Copyright (c) 2012 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 2, No 1, March 2012
ISSN (Online): 1694-0814
www.IJCSI.org 235
Table2: Statistical Measures of the proposed System for Video‐I using the proposed segmentation
approach
Mathews Kappa
Positive Negative False Correlation Coefficient
Measures TP TN FP FN Accuracy Predictive Predictive discovery Coefficient
value value(NPV) rate
Frame4 5370 94858 1125 1158 97.77 82.51 98.79 17.49 0.81 0.946997
Frame5 7569 97385 1266 1266 97.64 85.67 98.72 14.33 0.84 0.975712
Accuracy Comparison Graph‐I
100
Evaluation Metrics
98
96
94
92 Proposed
90
Prior Research
88
1 2 3 4
Frames
Figure 4: Accuracy Comparison graph‐I
The fig. 4 represents the efficiency of the proposed approach than the
The fig. 5 and fig. 6 represents the video respectively. The table-3 and table-4 represents the
segmentation results of the frames (Frame: 23, comparison statistical measures of the segmentation
Frame: 24, Frame: 25 and Frame: 65) using the prior of video-II using the proposed technique as well as
video segmentation research of the author and by the prior method and the fig.7 illustrates the
using the proposed segmentation technique corresponding accuracy comparison graph.
Figure 5: Segmented static and dynamic objects(prior research)
Copyright (c) 2012 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 2, No 1, March 2012
ISSN (Online): 1694-0814
www.IJCSI.org 236
Figure 6: Segmented static and dynamic objects (proposed method)
Copyright (c) 2012 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 2, No 1, March 2012
ISSN (Online): 1694-0814
www.IJCSI.org 237
BIOGRAPHY Dr.K. KUPPUSAMY
K.MAHESH
Prof. Dr K.KUPPUSAMY is working as an Associate
Mr. K. Mahesh is working as an Associate professor Professor in the Department of Computer Science
in the Department of Computer Science and and Engineering, Alagappa University, Karaikukdi,
Engineering, Alagappa University, Karaikudi, Tamilnadu, India. He has received his Ph.D in
Tamilnadu, India. He has received his M.Phil in Computer Science and Engineering from Alagappa
Computer Science from Bharathidasan University, University, Karaikudi, Tamilnadu in the year 2007. He
Tiruchirapalli, Tamilnadu in the year 2006. He has 21 has 22 years of teaching experience at PG level in
years of teaching experience at PG level in the field the field of Computer Science. He has published
of Computer Science. He has published many papers many papers in International Journals and presented
in international journals and presented his work in in the National and International conferences. His
national and international conferences. His areas of areas of research interests include
research interests include Video Segmentation, Information/Network Security, Algorithms, Neural
Video Processing, Clustering techniques, Image Networks, Fault Tolerant Computing, Software
processing. Engineering, Software Testing and Optimization
Techniques.
Copyright (c) 2012 International Journal of Computer Science Issues. All Rights Reserved.